Image-based Wear Debris Feature Recognition And Development Of Online Monitoring System

Modern industry is developing rapidly to the direction of intelligent and continuous,but wear has always been an important issue related to the service performance and service life of mechanical equipment,which makes the requirements for monitoring the operation status and fault analysis of equipment under actual working conditions higher and higher.Wear debris monitoring technology can determine the wear status of machinery through the analysis of wear debris,to evaluate the health status of mechanical equipment and components.Among them,the method of acquiring the image of wear debris and observing and analyzing the morphological characteristics,quantity,area,color of the wear debris in the oil to characterize the wear process can provide a complete understanding of the wear mechanism.The combination of computer image processing and wear debris monitoring technology has further promoted the ferrography technology to a new level.This thesis takes the wear debris image as the research object,firstly designs and develops the lubricating oil wear debris monitoring system that can obtain the wear debris image in real-time during the operation of mechanical equipment,then studies the algorithm that processes and identifies the features of the wear debris image,finally develops the planetary gearbox spalling failure test program to verify the system and the algorithm,realizes the health status monitoring of the planetary gearbox and analyzes its wear status.The main aspects of the study are as follows:（1）Firstly,according to the working principle of the lubricating oil wear debris monitoring system,the lubricating oil wear debris monitoring system was designed and developed from both hardware and software aspects,each hardware device of the lubricating oil wear debris monitoring system was selected,and each software interface was designed and the control program was written.（2）Taking the planetary gearbox as the test object,a healthy gear and three gear tests with different failure degrees were designed,and the health status was monitored from two aspects:online real-time monitoring and offline monitoring,and the wear debris image data was collected.（3）K-SVD dictionary learning algorithm was introduced to remove Gaussian noise on the image according to the characteristics of wear debris images,and the median filtering,mean filtering and Gaussian smoothing filtering algorithms were used to compare the denoising effects,and the subjective and objective indicators PSNR and SSIM were evaluated to prove the effectiveness of K-SVD algorithm for wear debris images.（4）Marked watershed segmentation has over-segmented or incompletely segmented many abnormally large wear debris.Therefore,the improved marked watershed segmentation algorithm（B-FSL）was studied according to the characteristics of the wear debris chain image,and the segmented wear debris were equivalent to ellipses to calculate their aspect ratio threshold to verify the accuracy of the B-FSL segmentation.（5）A method for recognition and classification of wear debris images based on the Res Net34 network and transfer learning is established.In order to enrich the data set of wear debris images,the data set was augmented,and the single grain of wear debris after the test image was denoised and segmented was input to the network for recognition and classification with an accuracy rate above 98%.Recorded and compared the classification results of the four sets of tests,and verify the severity of the spalling failure according to the type of wear debris generated in the tests.（6）Extracted the boundary energy（BE）characteristics of the wear debris according to the shape characteristics of the wear debris,analyzed the wear status and trend of the planetary gearbox under online monitoring,which proved that the BE features extracted in this thesis can provide early warning for the occurrence of planetary gearbox failures.The wear debris images collected under offline monitoring are processed by cropping,graying and inversion,and then their BE features were extracted,and the trend graph of BE features was compared with the trend graph of online monitoring,which verifies the effectiveness of online monitoring.The accuracy also confirmed the sensitivity of the BE feature to irregular large wear debris and played an early warning role compared to the wear debris concentration（IPCA）.